Retention of complex events is often thought to be reconstructive in nature, that is, the process of recall is one in which the rememberer retrieves whatever thematic structure he initially abstracted from the event and then uses general information (general in the sense that it is not tied to the particular initial event of interest), relevant to those themes to construct a logically plausible event. However, such a formulation does not enable us to account for situations in which memory for an event is highly accurate, or in which the original event is reproduced. An alternative model is proposed which specifies that among the contents abstracted from any complex event as it occurs are both generic or thematic attributes and detailed attributes. Over time the thematic attributes become more available than the detailed attributes and so are easily accessible at an attempted recall. The thematic attributes may then be used together with categorical information in semantic memory to produce a reconstruction of the original event. Or, they may be used together with other cues to retrieve the detailed attributes to produce a reproduction of the original event. The model then specifies a set of conditions under which one is likely to observe either reconstructive or reproductive recall. The three experiments here proposed test predictions generated by this model about the nature of recall. The experiments all follow a basic plan: the subject is exposed to complex prose material, either by hearing or by reading it; then, prior to recall, which may be immediate or after a delay of from 2 to 10 days, some treatment is delivered which according to the model should influence whether one observes either reproductive or reconstructive recall. Measures of retention include accuracy of recall and types of intrusions made by subjects.